Thrust 2: Field-Scale Assimilation and Validation

Lead: Tim Johnson

Goal: Achieve enhanced predictability of flow pathways at the testbed site by combining a basic understanding of processes and signatures with chemistry-controlled testbed circulation experiments and multi-physics sensing to directly inform high-performance PFLOTRAN simulations.

Photograph of a highly instrumented EGS testbed site
(Photo courtesy of Tim Johnson | Pacific Northwest National Laboratory)

Managing the performance of an enhanced geothermal system (EGS) reservoir over time depends on the ability to gather a variety of geophysical monitoring data on the fly and incorporate it into accurate predictive models. This thrust focuses on combining established geophysical sensing with novel advanced sensing modes to create real-time monitoring of EGS sites. While the current paradigm relies on independent analysis of disparate data streams, CUSSP efforts will target the development of integrated analysis using machine learning techniques to inform system modeling.

Task 2.1 Highly instrumented and controlled testbed experiments

 An important component of the CUSSP approach is the ability to conduct experiments in a highly instrumented and intensely monitored EGS testbed site. CUSSP will enhance the existing capabilities at our testbed site with additional monitoring wells and sensing instrumentation for higher fidelity information. The team will also inject proppant into the fracture network to access more realistic conditions found in EGS applications. This task will focus on designing and installing the new wellbore configuration and associated sensing equipment as well as performing specific chemistry-controlled circulation experiments. These experiments include gathering the baseline data on chemical reactivity and flows needed in Thrust 1 along with examining targeted changes to the system. The site will also be monitored for large-scale permeability through combined geophysical sensing, providing information that will be integrated into machine learning modeling.

Two photographs of borehole monitors
A) Installation of monitoring borehole string showing an electrical resistivity tomography (ERT) and multi-mode fiber on the outside of the casing. B) Completed borehole (prior to grouting) showing a continuous active-source seismic monitoring string (seismic sources and three-component geophones) on the inside of the casing. (Image courtesy of Tim Johnson | Pacific Northwest National Laboratory)
Plan view of the SURF testbed layout
Plan view of the SURF 4100 Level testbed layout, including existing instrumentation that will be leveraged by CUSSP and three proposed new boreholes. (Image by Jeff Burghardt | Pacific Northwest National Laboratory)

Task 2.2 High-performance thermal-hydrological-mechanical-chemical (THMC) and geophysics modeling

PFLOTRAN enables researchers to model subsurface flows and reactive transport as well as perform joint inversions of geophysical monitoring data. Members of the CUSSP team will enhance the capabilities of PFLOTRAN to encompass reactive flow through fracture networks. This task specifically targets the development of both forward simulation and rapid assimilation of observed data by PFLOTRAN. Through integration with other tasks and thrusts, the data obtained in laboratory and field experiments will be added to PFLOTRAN and used to evaluate the accuracy of simulations.

Figure showing simulation outputs
Discrete fracture network (left) and upscaled discrete fracture matrix (right) unstructured discretizations of a 3D fractured rock system produced by dfnWorks. (Image by Jeffrey Hyman from Sweeney, Matthew R., et al. 2020. Comp. Geosciences, 24, 293. | Colorado School of Mines and Los Alamos National Laboratory)

Task 2.3 Machine learning and joint inversion

With a robust understanding of the relevant physical and chemical processes enabled by other aspects of CUSSP research, this task will employ computational techniques to convert large amounts of data and established relational knowledge to identify differences between simulated outcomes and observations. The geophysical data gathered at the testbed site contains rich information about the overall system, but the complex relationships can make developing accurate predictions challenging. By employing machine learning models, developed in collaboration with domain experts and trained by PFLOTRAN simulations, and joint inversion techniques, CUSSP will effectively converge on the key underlying processes governing testbed behavior and thereby establish a new paradigm in EGS reservoir simulator development.